search

modelfoxdotdev / modelfox

ModelFox makes it easy to train, deploy, and monitor machine learning models.
Other
1.46k stars 63 forks source link

Selection of "positive" label? #69

Closed comunidadio closed 2 years ago

comunidadio commented 2 years ago

In binary classification, how does Tangram picks which label is considered as "positive" in tangram app reports (eg. Training Metrics) and for thresholding? any way to force the "positive" label according to the use case?

nitsky commented 2 years ago

The negative label is the first label and the positive label is the second. By default, the labels are sorted alphabetically. This works well for common values like false/true, no/yes, and negative/positive. You can use a configuration file to customize the order of the enum options. See the docs here: https://www.tangram.dev/docs/guides/train_with_custom_configuration. Let me know if you still need help!